Paperwork arrives, checked data leaves. Your staff handle the exceptions.
Grant applications, planning submissions, invoices and claims reach you as scans, PDFs, e-forms and emails. We build intake that sorts each one by type, pulls out the fields you need, tests them against your rules and puts only the doubtful ones in front of a person.
For Irish public bodies and companies that receive paperwork in English, Irish and other languages, and need to show where every figure came from.
The system takes in documents from all channels, works out their type, reads the fields and checks them. Anything with low confidence or a broken rule is placed in a queue for a person to review. Checked data passes to the business system, and every correction a reviewer makes is added to the test set.
Before and after
How the day changes for your staff
Today
With the system
Today
Today: Keying from attachments
Someone opens every scan, PDF or online form and types names, sums and reference numbers into the case file or finance system.
With the system
With the system: Fields come in ready to use
Each field is read together with the spot on the page it came from, so the reviewer has the value and its proof on one screen.
Today
Today: Mistakes surface too late
A missing page or a wrong figure only shows up weeks on, once a decision is appealed or a payment is returned.
With the system
With the system: Checks happen on arrival
Missing items, lookups against master data and your business rules are tested before a file is opened, while there is still time to go back to the applicant.
Today
Today: Easy and hard files cost the same
A tidy, complete invoice eats up as much staff time as an unclear one with half the details absent.
With the system
With the system: Staff time goes where it counts
Complete, clean documents flow through on their own. Staff work through a queue of doubtful and unusual items, each tagged with why it was stopped.
Today
Today: Nobody can say where a value came from
If an auditor queries a figure, the explanation rests on whoever typed it.
With the system
With the system: Each figure can be traced
For every field the log holds the source page, the value read, its confidence, the outcome of each rule and any change a reviewer made.
Workflow
The workflow, step by step
Seven steps and a single route for exceptions. Your process owners agree the thresholds and rules with us, and we tune them on your own paperwork before the system handles live work.
Illustrative workflow
7 steps · 1 human checkpoint
Legend
System
Person
Human checkpoint
Exception path
Document intelligence: illustrative workflow and review route
Read the diagram as text
The usual route goes from collection to identification, reading and checking, then on to the business system, and closes with improvements drawn from reviewer corrections.
Once checked, a document with any field under the confidence level or any broken rule is taken off the usual route and placed in the review queue. A person there accepts or corrects the values, and the document then returns to the route at the point where data passes to the business system.
That queue is where a person steps in: a doubtful value cannot reach the business system unless someone has reviewed it.
01
Step 1: Collect from all channelsSystem
Scanned post, emailed attachments, uploads from an online service and e-forms all go into a single intake queue, each with a file reference, details of the channel and a checksum. Structured EN 16931 e-invoices arriving over Peppol skip the reading stage and move directly to the checks.
02
Step 2: Separate and identifySystem
Bundles holding several documents are broken into their parts. Each part is labelled by type and by language, Irish or English included, and that label decides which field schema and rule set are used.
03
Step 3: Read the fieldsSystem
OCR that understands page layout reads the content, and a language model fills an agreed, strict schema. Each value comes with a confidence score and the place on the page where it was found.
04
Step 4: Test against your rulesSystem
Values are compared with master data such as customer and supplier records, VAT numbers and IBANs, with rules that link one field to another and with duplicate checks. A field that clears the agreed confidence level and every rule moves on without a person.
05
Step 5: Queue for reviewPerson
The reviewer gets the page and the values read from it on one screen, with the broken rule or doubtful field marked, and accepts or corrects each one.
Exception path: Low confidence or broken rule· Returns to step 6
Human checkpoint
No doubtful value goes into the business system unless a person has looked at it. A reviewer may also turn a document down and ask the sender for a fresh copy.
06
Step 6: Pass to the business systemSystem
Through its API, the checked data creates or updates the file in your case management system, CRM or ERP, with the original document attached and a retention class applied.
07
Step 7: Improve from correctionsSystem
Every correction is stored as a labelled example and added to the test set. Thresholds and prompts are only changed once a new version has passed that set.
Components
What we would put in place
01
Intake and channel connections
Links to shared mailboxes, scanners, online services and e-forms that bring each document into one queue with its file reference and metadata.
OCR that reads layout, combined with model-based reading into versioned schemas for each document type, giving a confidence score and page position for every field.
A labelled test set, gates before each release and dashboards showing the share handled without a person, why items were stopped and any drift in field accuracy.
Master data for customers, suppliers, VAT numbers and IBANs
Identity provider that assigns reviewer roles
Controls
Safeguards built in
Access boundaries
A reviewer sees only the files and document types their role permits. The reading services hold no permanent access to the business system; each hand-over runs under its own narrowly scoped service account.
Human review
A low confidence score or a broken rule sends the item to a person. Thresholds apply field by field rather than to a whole document, and we lower one only when the test set shows it is safe to do so. Case officers keep every decision on an application.
Auditability
The log keeps, for each field, its place in the source, the value read, the confidence, every rule outcome, the model and prompt versions, and any reviewer change with its reason code.
Data protection
Only the personal data you need is kept at intake, redaction is ready for access requests under the GDPR and the Data Protection Act 2018, and retention and deletion follow your records schedule. Processing remains in the EU region agreed with you.
AI transparency
We record each model and prompt version alongside its test results. When the output affects individuals, the use of AI goes into your AI register and is explained in the notice given to applicants, as the AI Act regime expects.
Measures
How we would judge it
These measures are settled with you in discovery, and we first take a baseline from the way you work today.
How we would judge it
Metric
Why it matters
How we would measure it
01Share handled without review
Why it mattersTells you, per document type, how much routine work no longer needs a person.
How we would measure itProportion of documents that clear every threshold and rule with no review, taken from the workflow logs.
02Exceptions and their causes
Why it mattersShows whether items are stopped by poor scans, absent information or rules that should be changed.
How we would measure itItems in the review queue broken down by reason code, channel and document type.
03Accuracy per field
Why it mattersThe quality figure that counts once data drives decisions and payments.
How we would measure itPrecision and recall for each field on an agreed labelled test set, run again with every release.
04Time per document
Why it mattersShows the effort saved and the effort that now sits in review.
How we would measure itTime from arrival to hand-over, plus hands-on review time per exception, read from queue timestamps.
No targets are set before a baseline exists.
Rollout
How we would introduce it
Phase 01
Discovery and baseline
Two or three document types and one channel, with today’s volumes and typical errors. We build a labelled test set from real documents with personal data removed.
Exit criteria
Fields, rules and quality bar signed off
Baseline recorded
Data protection impact assessment under way
Phase 02
Shadow pilot
The system reads and checks alongside the existing process; staff carry on as normal and compare results.
Exit criteria
Field accuracy reaches the agreed bar
A threshold fixed for each field
Reviewers at home on the workbench
Phase 03
Live use under supervision
Documents of the piloted types flow through without review, and a sample of those passed automatically is checked by staff.
Exit criteria
Samples reveal no unacceptable errors
Runbook and monitoring passed to your team
Rollback rehearsed
Phase 04
Extend
Further document types, channels and languages, Irish among them where you publish in Irish, each clearing the same test gate before it is switched on.
Grant and scheme applications with their supporting documents at State agencies, planning submissions at county councils, and FOI requests where records need redaction before release.
Delivery dockets, certificates of conformity and customs paperwork matched against purchase orders at manufacturing sites.
Illustrative example
Grant applications at a State agency
Situation
A State agency runs several grant schemes. Applications come in through an online service with supporting documents attached, by email, and as posted forms, some in Irish, and staff key the details into the grants system.
System
A single intake sorts the attachments, reads applicant details, amounts and scheme references, and checks them against the scheme rules and the applicant record in the grants system.
Human control
Files with a doubtful field, a missing document or a bank detail that does not match wait for a case officer. Only staff approve or refuse a grant, and payments are released from the finance system alone.
What we would measure
Share of applications handled without review by scheme and channel, reasons for stopping, and field accuracy on a labelled test set.
Ireland
Paper, PDF and Peppol, side by side
Under S.I. No. 258 of 2019, Irish contracting authorities must already receive and process EN 16931 e-invoices, and Ireland uses Peppol to carry them; and Revenue’s published VAT modernisation timeline plans the same structured format for domestic B2B invoices issued by large corporates from 1 November 2028. Structured invoices skip the reading stage entirely; scans and PDFs still need it, and both meet in one controlled intake.
Where read data feeds decisions about individuals, we design for the GDPR, the Data Protection Act 2018 and the Regulation of Artificial Intelligence Act 2026: processing on record, a person in review and logs your auditors and the DPC can follow.
That depends on your paperwork, so we measure it instead of promising a figure. We label a test set from your own documents with personal data removed and report precision and recall for each field. Reading models do get things wrong, and that is exactly why a doubtful field goes to a person and not directly into the business system.
02What about handwriting, weak scans, and documents in Irish?
Handwritten forms and weak scans can be read, but with lower confidence, so a larger share ends up in the review queue. Documents are sorted by language first, English, Irish or another EU language from a supplier abroad, so each is handled with the right schema and rule set.
03We already run OCR or a capture tool. Must it go?
Not as a rule. If your capture tool does its job, we add checking, review and hand-over around it, and bring in model-based reading only for document types where it pays off.
04Will the system decide on applications or claims?
No. It prepares and checks the data. Any decision with a legal or financial effect on a person stays with your case officers, in keeping with Article 22 of the GDPR and your own policies.
05Where do the documents get processed and kept?
In whichever cloud region or data centre you select, inside the EU. Every processing step is written down, any outside model service included, so your data protection officer can review it before go-live.
06Where should we start?
With one high-volume document type and clear rules, for example supplier invoices or one grant application form. That tests the workflow, the review screen and the measures before more types come in. See how we work.
Show us your paperwork
Pick two or three document types and the systems they land in. We will sketch what a first release could look like, in English and Irish where you need both.